Date of Award
Master of Science (MS)
Animal and Veterinary Sciences
Dr. Heather Dunn
Dr. Charles Rosenkrans
Dr. Federico Iuricich
Dr. Nina Hubig
Breast cancer continues to be the most diagnosed cancer in the United States, affecting one in eight women and the second leading cause of cancer death. Despite advancements in treatment therapies and early detection strategies, cancer health disparities remain for racial and ethnic minority groups with African American women having the highest overall breast cancer mortality rate. Current research suggests biological differences in the extracellular matrix (ECM) among racial and ethnic groups, thus contributing to the incidence and mortality in African Americans.
Our research is focused on investigating racial disparities in breast cancer utilizing machine learning (ML) and artificial intelligence (AI) techniques to examine breast cancer morphologies between racial groups. To evaluate morphological patterns of the ECM between racial groups, we developed an interpretable AI framework that can differentiate between African American and Caucasian histology images at a high level of accuracy. Our machine learning classifier was designed with a model provided by PyTorch’s torchvision package, ResNet50. Our model resulted in a classification accuracy of 92.1% when evaluating 20% of our breast cancer histology images. Results were analyzed using local interpretable model-agnostic explanations (LIME) and saliency mapping for model interpretability. LIME highlighted regions of the image deemed significant for output classification. Saliency mapping-colored pixels in the input image were responsible for the output classification. Both, saliency mapping and LIME highlighted regions of the ECM and the interface of tumor tissue, indicating machine-detected differences between our two racial groups.
We created an AI model using a cloud-based platform, Aiforia (Aiforia Inc., Cambridge, MA), to evaluate protein expression of human breast cancer biopsy samples from African American and Caucasian women. Tissues were immunohistochemically (IHC) stained with monoclonal antibodies, Vimentin, CD29/β1 integrin, and α-SMA to identify cancer-associated fibroblasts (CAF). Our AI model identified the total tissue surface area and quantified the total number of positive CAFs. Statistical analyses showed Caucasian women having a higher number CAFs in tissues stained for CD29 and α-SMA than African American women. To confirm genetic expression differences between African American and Caucasian breast cancer patients, laser micro-dissection (LMD), ribonucleic acid (RNA) extraction, and sequencing techniques were performed.
Stone, Amber, "An Explainable Artificial Intelligence Approach to Differentiate Breast Cancer Morphologies Between Racial Groups" (2022). All Theses. 3826.